Fachgruppe Sprachwissenschaft

Universität Konstanz

Arbeitspapier 78

Turkish Relative Participles.A Reanalysis in Categorial Grammar.

Oktober 1996

Klaus von Heusinger

Turkish Relative Participles.A Reanalysis in Categorial Grammar.*

Klaus von Heusinger, FG Sprachwissenschaft

Universität Konstanz Postfach 5560 D 185, 78434 Konstanz

email:klaus.heusinger@uni-konstanz.de

September 1996

* The work on this article was supported by the German Science Foundation (EG 54/1-2).Furthermore, I like to thank Andreas Ulrich for discussing this paper with me and JeniferCole for giving me very helpful comments for revising the paper. All remainingshortcomings are mine.

- 2 -

CONTENTS:

1. INTRODUCTION.......................................................................................................................................3

2. CATEGORIAL GRAMMAR AND LINGUISTIC THEORY.............................................................................4

2.1 Classical Categorial Grammar.........................................................................................................5

2.2 Generalized Categorial Grammars...................................................................................................6

2.3 Morphology and Bracketing Paradoxes..........................................................................................10

3. TURKISH AS A CATEGORIAL LANGUAGE.............................................................................................15

3.1 Nominals and Nominal Constructions.............................................................................................15

3.2 The Izafet-Construction..................................................................................................................18

3.3 The Verbal Complex.......................................................................................................................19

3.4 Participles......................................................................................................................................20

3.5 Verbal Nouns..................................................................................................................................23

4. TURKISH RELATIVE CLAUSES..............................................................................................................25

4.1 A Double Bracketing Paradox........................................................................................................26

4.2 The Bah¦bozuk or ‘Broken Head’ construction................................................................................28

4.3 Flat Reconstruction of Embedded Sentences...................................................................................30

BIBLIOGRAPHY.........................................................................................................................................31

- 3 -

1. IntroductionUnderhill (1972) analyzes the two types of relative construction in Turkish. Onetype is applied if the subject of the embedded clause is relativized, whereas thesecond construction covers all the other cases. In this paper, I try to give atransparent reanalysis of the suffix -digi which forms relative constructions ofthe second type. The relative clause in (2) is derived from the simple sentence in(1) by modifying the verbal stem bekle with the participle suffix -dig and thepossessive suffix -i. The subject of the relative clause is realized by the genitivemodifier kardehimin of the participle bekledigi:

(1) kardeh-im misafir-i bekli-yorbrother-pos.1.sg guest-acc expect-pres‘My brother expects the guest.’

(2) kardeh-im-in bekle-dig-i misafirbrother-pos.1.sg-gen expect-part-pos.3.sg guest‘the-waiting-of-my-brother-guest’ = ‘guest that my brother expects’

On one hand the participial suffix -dig forms a participle from a verbal stem,which modifies the head noun misafir. On the other hand the possessive suffix -irefers to the genitive kardeh-im-in, which is the subject of the relative clause.This causes a double bracketing paradox: The participial suffix -dig refers to thehead noun misafir while the possessive suffix refers back to the genitive kardeh-im-in crossing the attributive participle: [1 kardeh-im-in bekle-[2 dig-i]1 misafir]2.These complex functional relations can only be explained if themorphology/syntax border is made more transparent than it is assumed in mosttheories about syntax. I argue with special reference to the relative suffix digithat a categorial analysis gives new insights into the syntax-morphologyinterface. It will be shown that Turkish relative clause constructions can betraced back to general syntactic and morphological rules of Turkish.

This paper is organized as follows: In section 2, I give a short outline ofcategorial grammars in linguistic theory. The basic ideas are introduced by theexample of the classical formalism due to Ajdukiewicz. Then, generalizedcategorial grammars are discussed, which are able to describe more complexphenomena, like bracketing paradoxa. In section 3, I sketch some essentialconstructions of Turkish which are crucial to the analysis of relative clauseconstructions: nominal constructions, genitive groups and participles. Turkishnouns can be used as substantives, attributes or predicates. The Turkish genitive

- 4 -

group or definite izafet-construction is marked by an agreement feature betweenthe modifying genitive and the head noun. Participial constructions andnominalizations play an important role in Turkish because they are the means toform subordinate clauses. The categorial description of these three grammaticalconstructions yields a transparent analysis of Turkish relative clauses andrelative clause suffixes in section 4. It can be shown that the reanalysis of thefunction of the suffix -digi does not only solve the double bracketing paradox,but also provides a transparent composition of its lexicalized meaning. Finally,the analysis is applied to the construction that Lewis (1967) called ‘bah¦bozuk’(‘broken head’) like oglan-¦n mekteb-in-e git-tigi adam ‘The man, to whoseschool the son goes’.

2. Categorial Grammar and Linguistic Theory

Categorial grammars are the oldest explicit formal systems for the description ofsentence structure. They are based on three main ideas: Husserl’s notion of‘Bedeutungskategorie’ (‘category of meaning’), Frege’s reasoning on thefunctionality of language and Russell’s formulation of the stratification oflanguage. The first categorial grammar unifying the three components wasformulated by Le,niewski (1929) and elaborated by his pupil Ajdukiewicz(1935). Bar-Hillel (1953) applied this logical and formal approach to the notionof constituents used in the structuralist tradition and coined the notion ofcategorial grammar. Later he followed Chomsky in doubting the adequacy ofcategorial grammars, which are a weakly equivalent to constituent structuregrammars. This caused a lack of interest in the linguistic application ofcategorial grammars during the following decades. However, categorialgrammars were investigated by mathematicians like Lambek (1958),philosophers like Geach (1972) or semanticists like Cresswell (1973). Theywere also applied by Montague (1974) as the syntactic equivalent to thesemantic component supporting the central principle of compositionality.

Not until the eighties were categorial grammars back in linguistic discussion(cf. Buszkowski et al. 1988 and Oehrle et al. 1988). There are the followingreasons among others: The growing research in automatic language processingcaused the interest in syntactic languages that entail their semanticsunambiguously. The shift of grammatical complexity from the syntax to the

- 5 -

lexicon leads to a recollection of very simple syntactic mechanisms, which hadbeen investigated by categorial grammar for a long time. Finally, extensions ofthe theory like polymorph categories, complex feature structures and the rule ofunification have opened new applications (cf. Haddock et al. 1987 or Klein &van Benthem 1988).

2.1 Classical Categorial Grammar

The concept of ‘classical categorial grammar’ denotes a formal system goingback to Ajdukiewicz and Bar-Hillel. It is characterized by the following threepoints:

(i) There exists a strict parallelism between syntactic and semanticcategories.

(ii) The only syntactic operation permitted is concatenation of a functorwith its argument.

(iii) Syntactic concatenation is interpreted semantically as functionalapplication, i.e., as the application of the meaning of the functor to themeaning of the argument.

Classical categorial grammar uses only two basic categories, n for name and sfor sentence. All other linguistic categories are functors that receive morecomplex indices constructed out of the basic ones. Table (4) shows apreliminary assignment of indices to linguistic expressions in a simple grammarwith forward application.

(4) category categorial index language expression

sentence s the rose bloomsname n Socratescommon name s/n rose, bear, ducknominal, term s/(s/n) nobody, the roseintransitive verb s/n bloom, laughmodal verb (s/n)/(s/n) can, mustattributive adjective (s/n)/(s/n) big, smallarticle (s/(s/n))/(s/n) the, a(n)adverb (s/n)/(s/n) strongly, loudlysentence operator s/s not, necessary that

- 6 -

The formalism works with one basic syntactic rule, namely concatenation orforward application. A functor expression A can be applied to an argumentexpression B if A has the categorial index X/Y (X over Y) and B the categorialindex Y. The resulting index of the concatenated expressions A^B is X.Concatenation is interpreted as functional application, i.e. as the application ofthe meaning of the functor category to the meaning of the argument category.

(5) Concatenation or forward application: X/Y Y → Xf x → f(x)

A categorial grammar with forward application and the indices given in (4) canalready analyze fragments of English as illustrated in (6) and (7). Thederivations are represented in ‘flat’ structures, contrary to the hierarchical treesof constituent structure grammars. The categories are concatenated from left toright applying a functor to the argument on its right.

(6) the rose blooms(s/(s/n))/(s/n) s/n s/n

s/(s/n) s

(7) the little bear could loudly laugh(s/(s/n))/(s/n) (s/n)/(s/n) s/n (s/n)/(s/n) (s/n)/(s/n) s/n s/n s/n

s/(s/n) s/ns

However, the range of application of classical categorial grammar is restricted,because it can only give derivations for structures that have a functor expressionoccurring before its argument.

2.2 Generalized Categorial Grammars

Generalized categorial grammars use additional rules in order to describe awider range of data: backward application, functional composition, typechange rules and unification. The rule of backward application (8) gives agreater flexibility in the directionality of concatenation. Backwardapplication permits a functor to take the arguments on its left side, like in

- 7 -

(9a). The meaning of the intransitive verb walk in (9a) is applied to themeaning of the name Mary: walk'(Mary').1 The basic word order in Englishcan be described by a categorial grammar combining forward and backwardapplication, as illustrated in (9b).

(8) backward application Y Y\X → Xx f → f(x)

(9a) Mary walksn n\s

s

(9b) Mary beats Harryn (n\s)/n n n\s

s

Backward application as well as forward application is interpreted as theoperation of applying the meaning of the functor to the meaning of theargument. Since the two rules of application are not semantically distinguished,they are summarized as functional application (= FA). We will mark thedirection of application with a lower-case letter (FAf vs. FAb). A categorialgrammar with only one direction of application is called unidirectional, and agrammar with both applications bidirectional.

Functional composition combines two functor categories if the type of theargument that the first functor demands is equal to the resulting value of thesecond functor. Functional composition is interpreted as the semantic operationthat assigns the function λx [f(g(x)] of type X/Z to the composition of a functionf of type X/Y and a function g of type Y/Z. There are two directional versions:

(10) functional composition (FC)

(i)forward (FCf): X/Y Y/Z → X/Zf g→ λx [f(g(x))]

(ii) backward (FCb): X\Y Y\Z → X\Zg f → λx [f(g(x))]

1 The meaning of predicates and names is not analyzed further. Following Montague (1974),they are represented by the ‘'’ (prime) on the corresponding constants for predicates andindividuals.

- 8 -

The application of functional composition can be illustrated by example (11).The adverbial all year is categorized as a sentence modifier of type s\s. In (11a)it modifies the sentence no lady wept. However, in (11b) it combines first withthe predicate wept yielding a new predicate wept all year, which is the argumentof the quantifier phrase no lady. The construction wept all year is licensed byfunctional composition, which constructs the functor of type s\s with anotherfunctor of type n\s yielding the complex functor of type n\s. The order ofapplication causes a difference in meaning. Derivation (11a) means that all theyear, there is no lady who wept, whereas derivation (11b) expresses that there isno lady who wept the whole year:

(11a) No lady wept all years/(n\s) n\s FAf s\s

s FAb s ‘all year: no lady wept’

(11b) No lady wept all years/(n\s) n\s s\s FCb

n\s FAf

s ‘no lady: wept all year’

This set of syntactic rules is completed by type changing rules, with theGeachian rule being the only one discussed here. According to this rule, theindex of a functor category may be expanded to a more complex index:

(12) Geachian rule (G)

X/Y → (X/Z)/(Y/Z) or Y\X → (Z\Y)\(Z\X)f → λgλx [f(g(x))] with g of type Y/Z and x of type Z

The Geachian rule is closely related to functional composition, which can beillustrated by comparing (11b) with (13). In (11b) the two functor categories arecombined by functional composition, whereas in (13) the functor all year oftype s\s is first extended to a more complex functor of type (n\s)\(n\s), whichthen is concatenated with the predicate wept of type n\s by functionalapplication.

- 9 -

(13) No lady wept all years/(n\s) n\s s\s

| G with Z = n

(n\s)\(n\s) FAb

n\s FAf

s

Another extension of the grammar concerns polymorph categories andunification (Uszkoreit 1986). An adverb can be assigned the different categoryindices s/s, (n\s)/(s\n), ((n\(n\s))/((n\(n\s)) or (n\(n\(n\s)))/(n\(n\(n\s))) modifyinga sentence, a transitive, an intransitive or a bitransitive verb, respectively.Although these category indices could be derived by the Geachian rule from thebasic type s/s, unspecified or polymorph category indices do the same jobwithout derivations. An adverb is assigned the underspecified index (x\s)/(x\s)with x being a variable whose value is determined by unification.

The same holds for the conjunction and whose function is to combine twoexpressions of the same kind to a complex expression of that kind. This can begenerally captured by the underspecified index x\x/x.2 In (14) and combines twoexpressions of type n, and in (15) two sentences of type s. The adequate index isderived by unification, which assigns the index n\n/n for the coordination ofnames and s\s/s for the conjunction of sentences:

(14) Dädalus and Ikarus are flyingn x\x/x n n\s

| U with x = n

n\n/n n

s

(15) Dädalus flies to Crete and Ikarus plunges in the oceans x\x/x s

| U with x = s

s\s/s s

2 The unspecified category x\x/x is given here without any brackets, and this means, that it isnot important, whether the functor expression takes the first argument from the left or fromthe right or from both sides at the same time.

- 10 -

Summarizing, we can state the following category indices and someabbreviations for complex functor indices. N replaces the index n\s for commonnouns, while IV substitutes the category n\s for intransitive verbs and TV standsfor n\(n\s) for transitive verbs. Other abbreviations are used according to table(16):

(16) category category index abbreviation

common nouns n\s N(Montague: CN)NP, term s/(n\s) NP (Montague: T)verb intransitive n\s IVverb transitive n\(n\s) TVmodal verb (x\s)/(x\s) IV/IV or TV/TVadverb (x\s)/(x\s) IV/IV or TV/TVattributive adjective (n\s)/(n\s) Adj or N/Narticle (s/(n\s))/(n\s) NP/Nprepositional phrase (x\s)/(x\s) PPconjunction x\x/x conj.

2.3 Morphology and Bracketing Paradoxes

Bidirectional categorial grammars without any type change rules can be used forthe description of large fragments of morphology. Morphological derivations areanalyzed by assigning a lexical category to the stem, and functor categories toderivational affixes. In (17) law has the lexical category N for common nouns.The derivational morpheme -ful derives an adjective from a noun, and therefore,is of the type N\Adj. The negation suffix un- is of the type Adj/Adj, for derivingan adjective from an adjective. Finally, the suffix -ness can be categorized asAdj\N, because it forms a noun from an adjective. In (17) the adjective lawful isderived from the nominal stem law. This is then transformed to the adjectiveunlawful, which finally leads to the noun unlawfulness. The semantic analysis iscorrespondingly interpreted as the subsequent functional application of thederivational morphemes: NOM(NEG(ADJ(law'))).

- 11 -

(17) un- law- ful-nessAdj/Adj N N\Adj Adj\N ADJ(law')) Adj NEG(ADJ(law'))

Adj NOM(NEG(ADJ(law')))N

The step-by-step derivation in (17) is morphologically and semanticallytransparent by attesting a linguistic form for every stage in the derivation.Furthermore, the order of the derivation is directly related to the semanticcomposition of the meaning which is constructed out of the meanings of thelexical stem and the meanings of the derivational affixes. However, in othercases there is no such a clear structuring. On one hand, many derivations are nottransparent anymore, and on the other hand, the morpho-syntactic order ofderivation deviates from the semantic composition. The cases in question are theso called ‘bracketing paradoxes’ (see Williams 1981). Moortgat (1988)investigates bracketing paradoxes in a categorial framework.

In (18) the adjective sprachlos (‘speechless’) is composed of the nominalstem sprach (‘speech’) and the suffix -los. (‘-less’) The noun Sprachlosigkeit(‘speechlessness’) is derived from this adjective. However, there exists nointermediate form sprachlosig. Therefore, Moortgat (1988, 325) assumes alexicalized suffix -igkeit of type Adj\N, which is an allomorph to -heit or -keit.

(18) Sprach- los- [ig- keit]speech-less- nessN N\Adj Adj\N ADJ+WITHOUT(speech')

Adj NOM(ADJ+WITHOUT(speech'))N

Preliminary stages of this phenomenon can be synchronically observed in theform of ‘reanalysis’. In the morphologically motivated simple analysis in (19),the nomen agentis Spieler is derived from the verbal stem spiel. The suffix -informs the female form Spielerin. However, in the reanalysis the two suffixes arefirst connected by functional composition, and then they are applied as acomplex suffix to the verbal stem. It is interpreted as follows: The functionFEM, which constructs a female noun from a male noun, is applied to thefunction AGENS. Though the semantics of both derivations is identical, thereanalysis provides a proper meaning of the complex suffix -erin. A certaintendency towards lexicalization of this suffix -erin can be seen in the derivation

- 12 -

Gebärerin (female that gives birth to somebody) that obviously is not derivedfrom the substantive Gebärer.

(19) simple N FAb FEM(AGENS(play'))analysis N FAb AGENS(play')

V V\N N\NSpiel- er- inV V\N N\N FCb λP FEM(AGENS(P))

Reanalyse V\N FAb λP FEM(AGENS(P)) play'N = FEM(AGENS(play'))

Another prominent example of reanalysis is discussed by Di Sciullo & Williams(1987, 71ff.). Here, the word hydroelectricity is represented by the structure in(20) for purely morphological reasons, although it is derived from hydroelectric.Its meaning can be compositionally built up from the meaning of the wordhydroelectric and the meaning of the affix -ity. But the morphological structuredoesn’t correspond with this semantic composition, because -ity is an affix ofclass I, which can not be affixed to a word that already begins with an affix likehydro belonging to class II (Selkirk 1982):

(20) N

N/N N

hydro Adj Adj\N

electric ity

Di Sciullo & Williams (1987, 73) show that “the bracketing paradoxes spansyntax and morphology. For example, the syntactic phrases transformationalgrammar and transformational grammarian are ‘related’ exactly ashydroelectricity and hydroelectric are:”

- 13 -

(21a) NP

Adj NP

transformational N af

grammar ian

(21b) NP

Adj N

transformational grammar

Following Williams (1981), Di Sciullo & Williams (1987, 72) define the‘relation’ as a configuration of lexical units specified by the followingdefinition:

(22) X is related to Y if X can be got from Y by substituting for a head ofY, including substituting 0 for a head of Y.

This definition is rather general and concerns only the interpretation ofexpressions, but not the derivation. The following example shows how this puresemantic ‘relation’ can influence the syntactic description. The relation betweenthe noun relation and the noun relationship can be described in the followingway: The derived noun subcategorizes for the same prepositional phrase as thebase relation. However, according to morpho-syntactic principles, thenominalizing suffix -ship is first applied to the noun, and the resulting noun isthen applied to the PP. In contrast, the semantic interpretation process firstmerges the expression relation and the PP, and then nominalizes the whole:

Adj/PP Adj\N PPrelation- ship to NP

N

Adj\N N

morphological

semantic

Adj/PP Adj\N PP

(23) NP

Adj

A ‘disharmonic’ functional composition must be formulated in order to composerelation and ship. Differing from the harmonic functional composition rules in(10), the disharmonic or ‘mixed composition’ rule (Moortgat 1988, 320)

- 14 -

combines two functors of different directionality. This composition ruleadequately describes the semantic relations in (23) as the derivation (23a).

(24) disharmonic functional composition

(FCdish) X/Y X\Z → Z/Yg f → λx [f(g(x))]

(23a) relation- ship with NPN/PP N\N FCdish PP

N/PP λx (NOM(relation'(x))N λx (NOM(relation'(x)) (to_NP’)

= (NOM(relation'(to_NP’))

It was shown that the simple categories of categorial grammar give a greatertransparency of grammatical constructions between morphology and syntax.This constructive possibility of analyzing structures is applied to Turkishrelative clause constructions, which are traditionally considered as being verycomplex. But before a description of this heterogeneous structure is given, thenext section provides a short introduction to the grammatical data of Turkishthat are important to this investigation.

- 15 -

3. Turkish as a Categorial Language

Turkish is an agglutinative language with suffixes and SOV word order.Modifying elements like genitives, attributive adjectives, numerals, adverbs andrelative clauses precede the modified element.3 In sentence (25), the subjectressam stands in initial and the predicate gösterdi in final position. In thecanonical word order, the indirect object precedes the direct object, which islocated just before the verb. The verb, as head of the sentence, demands itsarguments of type NP or s/(n\s) from right to left. The adverbial complementgeçen hafta modifies a constituent of type IV (= intransitive verb). Thecomplement itself is constructed according to the principle ‘modifier precedesmodified’: the modifying element geçen precedes the modified element hafta.The derivation of sentence (25) can be described in a bidirectional categorialgrammar with functional application as being the only construction rule:

(25) ressam geçen hafta bize resim-ler-i-ni göster-diartist last week 1.pl.dat picture-pl-pos.3.sg-acc show-di.pastNP IV/IV NP NP NP\(NP\(NP\s))

= NP\(NP\s) (NP\s)/(NP\s) NP\s

NP\ss

‘Last week the artist showed his pictures to us.’

3.1 Nominals and Nominal Constructions

Turkish nominals are traditionally categorized into nouns and adjectives,although there is no clear cut distinction (Lewis 1967, 53):

The dividing line between noun and adjective is a thin one, but still worth drawing.If we take as the criterion of a noun the permissibility of using the plural, case, andpersonal suffixes after it, or the indefinite article bir before it, very few of thewords classed as adjectives in the dictionary will be excluded. büyük ‘big’, ‘old’,büyüklerim ‘my elders’ (...).

3 See the grammars of Kissling (1960) and Lewis (1967) for a comprehensive introducion,and Underhill (1986) for a short overview.

- 16 -

I assume that nominals can take three different modes in a sentence: Theyfunction either as substantives or modifiers or predicates. The function of asubstantive is marked by the substantival suffixes. The substantive, or the headnoun of an NP, stands in phrase-final position, i.e., all modifying elements likeadjectival attributes, modifying genitives or relative clauses are located beforeit, while the substantival suffixes follow it. A substantive consists of a stem andthe plural suffix -lEr,4 a possessive suffix, and a case suffix. Binding consonantsbridge two vowels, e.g., the n in resim-ler-i-ni is put between the possessivesuffix i and the accusative suffix i. The absolute, i.e., the nominative and theindefinite (or nonspecific) accusative, is not overtly marked by a case suffix.Both plural suffixes and possessive suffixes are optional elements:

(26) stem (+ plural) (+ possessive) + case

(27) ressam-Ø güzel resim-ler-i–ni göster-diartist-nom beautiful picture-pl-pos.3.sg-acc show-di.past‘The artist showed his beautiful pictures.’

In the second function, nominals can act as modifiers. However, there issome lexical restriction which divides nominals into two lexical classes: nounsand adjectives. The latter can undergo comparison and modify nouns asattributive adjectives. The former can modify another noun only by means of anizafet-construction (see section 3.2. below).

Thirdly, nominals are predicates in nominal sentences. The differencebetween the attributive and predicative use of adjectives can only be inferredfrom the position in the sentence, as is illustrated by the contrast in (28):

(28) güzel resiml-er vs. resiml-er güzel (dür)beautiful picture-pl vs. picture-pl beautiful (copula)‘beautiful pictures’ ‘The pictures are beautiful.’

Nouns are either of the type s/(n\s) being NPs, or of the type n\s beingpredicates. Due to their semantic characterization as properties, they are

4 Following a general convention, upper case letters represent underspecified phonemes.They are subject to vowel harmony and other morpho-phonemic processes. The I stands forthe fourfold vowel harmony, according to which the vowel depends in frontness androundness on the preceding vowel, i.e., it is realized as i, ¦, ü or u. E stands for the binaryvowel harmony, which predicts e or a according to the frontness of the preceding vowel.

- 17 -

assigned the lexical type n\s. The nominal type s/(n\s) is derived by theapplication of case suffixes, which get the same category index as the articles inEnglish (cf. (16)). They change an expression of the type n\s to an expression ofthe type s/(n\s). Possessive and plural suffixes are of type (n\s)/(n\s).

(29) substantival suffixes

suffix type abbreviationplural (n\s)\(n\s) N\Npossessive (n\s)\(n\s) N\Ncase (n\s)\(s/(n\s)) N\NP

(27a) ressam-Ø güzel resim-ler-i-ni göster-diartist-nom beautiful picture-pl-pos.3.sg-acc show-di.pastN N\NP N/N N N\N N\N N\NP NP\(NP\s)

NP N NP

NP\ss

Adjectives, on the other hand, receive the underspecified category index(n\s)/x, which is flexible enough to describe the contrast between theirattributive and predicative function. For the attributive use the variable x isinstantiated by n\s and results in the index (n\s)/(n\s) or N/N, whereas in thepredicative use the x is instantiated by Ø, which yields the index n\s. Thesubstantival function is determined by case suffixes, whose category index,however, must be modified: It takes an expression with the polymorph index(n\s)\x as its argument. The contrast between the predicative and the attributivemode in (28) can be described as follows:

(30) the modes of adjectives

predicative attributive substantivallexical type (n\s)/x (n\s)/x (n\s)/xsyntactic constr. predicative attributive + case suffixoperation U with x = Ø U with x = n\s ((n\s)/x)\s/(n\s)resulting type n\s = IV (n\s)/(n\s) = N/N s/(n\s) = NP

- 18 -

(28a) güzel resim-ler-Ø resim-ler-Ø güzelbeautiful picture-pl-nom picture-pl-nom beautifulN/x N N\N N\(NP) N N\N N\NP (n\s)/x | U x = N NP U x = Ø |N/N = s/(n\s) n\s

N sNP

3.2 The Izafet-Construction

Nouns can modify other nouns by means of a genitive construction or izafet-construction, like in (31). The agreement in person and number between thegenitive and the head noun is marked in the possessive suffix adjoined to thehead noun.5 In the categorial analysis, the genitive suffix is taken as a functorcategory that is applied to a common noun of type N and yields an NP with thesub-category gen. Subcategories are drawn in square brackets behind thecorresponding main category. The possessive suffix is taken as a functor, whichchanges a N to an NP that lacks a genitive phrase of the type NP[gen]. Izafet-constructions can be embedded almost without restriction as illustrated in (32).The izafet-group adam-¦n1 baba-s¦1-n¦n2 modifies the head noun ev–i2–ni, whichitself stands in the accusative case:

5 The constructions discussed here belong to the group of definite izafet-constructions. In theindefinite izafet-construction, the modifying noun recieves no genitive case, and thegrammatical relation is rather ‘qualifying’. The difference can be illustrated with thefollowing pair (cf. Lewis 1967, 41ff.).

(i) universite-nin profesörler-i vs. universite profesörler-i‘the professors of the university’ vs. ‘university professors’

- 19 -

NP[gen] NP[gen]\N

(31) N

uzman -¦n raport -¦ ‘report of the expert’expert -gen report -pos.3.sgN N\NP[gen] N N\(NP[gen]\N)

NP[gen] NP[gen]\NN

(32) [[adam-¦n baba-s¦ ]izafet1 -n¦n ev -i ]izafet2 -ni al-d¦-m¦z

man-gen father-pos -gen house-pos -acc by-past-1.pl

N N\NP[gen] N N\(NP[gen]\N) N\NP[gen] N N\(NP[gen]\N) N\NP[acc] NP\s

NP[gen] (NP[gen]\N) (NP[gen]\N)

N

NP[gen]

N

NP[acc]

s

‘We bought the [house of the [father of the man] ]’

3.3 The Verbal Complex

Turkish verbs consist of a stem and a variety of suffixes: The base or extendedstem is composed of the stem and optional derivational suffixes, which formdifferent diatheses, and an optional suffix for negation. The base is completedwith tense and aspect suffixes and a personal ending. The personal ending of the3rd person can be deleted (cf. Lewis 1967, 153):

(33) stem (+ derivation) (+ negation) + tense/aspect + personal endingbase or extended stem

(33a) tan¦-h-t¦r-¦l-ma-d¦-larknow-rec-caus-pas-neg-di.past-pl‘they were not caused to know each other’= ‘they were not introduced to one another’

- 20 -

(33b) ac¦-n-d¦r-¦l-ma-d¦-kfeel pain-refl-caus-pas-neg-di.past-1.pl‘we were not caused to feel pain in ourselves’= ‘we were not made to grieve’

Nominals can be derived from all verbs. They are categorized as substantives oradjectives according to their construction and syntactic function. In thefollowing, I distinguish between participles and verbal nouns. Both groups aresemantically different, because participles express a property whereas verbalnouns denote facts or propositions. Both groups show verbal subcategorizationfor the internal arguments and nominal subcategorization for the subject, whichis constructed as genitive. Participles or deverbal aare mostly constructedattributively, but they can also be used as substantives or predicates. Verbalnouns are normally used as substantives and they stand for subordinatedclauses. The relative clauses ending in -DIgI take an interesting intermediateposition, because they are constructed like verbal nouns on the one hand, butcan be used attributively on the other hand.

3.4 Participles

The participles are formed by a set of suffixes adjoined to the base. Theypartially correspond to the finite tense/aspect forms as shown in (34). Thefollowing analysis concentrates on the two suffixes that play a primary rolein building relative clauses: The En-suffix forms the subject participle (SP),whereas the object participle (OP) is built by the DIK-participle and thepossessive suffix -I.6

6 The aorist participle -Er and the past participle -mih behave similar to the present participle-En, while the future participle -EcEK is constructed in the same way as the DIK-participle.The notions ‘subject participle’ and ‘object participle’ are due to Underhill (1972) andHankamer & Knecht (1976). They describe the function of the participle in the derivation ofthe relative clause (cf. section 4).

- 21 -

(34) tense/aspect finite verbal form participle form of suffix

present geli-yor-um gel-en Enaorist gel-er-im gel-er Ermih.past gel-mih-im gel-mih mihdi.past gel-di-m gel-dik DIK7

future gel-eceg-im gel-ecek EcEK

The subject participle is derived from the verbal stem with the suffix -(y)En.Hence the participle bekli-yen ‘waiting’ can be obtained from the verbal stembekle- ‘wait’. We can also form the participle bekle-mi-yen ‘not waiting’ fromthe corresponding base bekle-me ‘not-wait’. The participle inherits thesubcategorization from the verbal base: biz-i (acc) bekliyen ‘waiting for us’.Participles that are built in this way can function as attributives or substantives.In (36), the participle suffix -En changes an intransitive verb to an adjective; in(37) the accusative case suffix -i forms a nominal from the participial form.There is no independent predicative use, which is rather covered by thecorresponding verbal form as in (35):

(35) misafir biz-i bekli-yorguest we-accwait-pres ‘The guest is waiting for us.’

(36) biz-i bekli-yen misafirwe-acc wait-SP guest ‘guest that is waiting for us’

(37) biz-i bekli-yen-i gör-dü-mwe-acc wait-SP-acc see-di.past-1.sg‘I saw the one waiting for us.’

The participial suffix -En receives the type IV\(N/N) due to its function offorming an (attributive) adjective from an intransitive VP. The substantival useof such a verbal adjective is derived through the application of the case suffix oftype (N/x)\NP.

7 The K of the DIK- and EcEK-suffixes changes intervocalically to ‘yumuhak ge’, i.e. a ‘softG’, which is realized, depending on the dialect, as vocalic lengthening or as a glide.

- 22 -

(36a) biz-i bekli -yen misafirwe-acc wait -SP guestNP NP\(NP\s) IV\(N/N) N

NP\s = IV N/N

N

(37a) bizi bekli -yen -i gör-dü-mwe-acc wait -SP acc see-di.past-1.sgNP NP\(NP\s) IV\(N/N) (N/x)\NP NP\s

NP\s = IV |N/N | U with x = N

(N/N)\NPNP

s

The distribution of the DIK-participle is restricted. It either exists in lexicalizedforms like in (38) or it is constructed with negated verbal stems like in (39):

(38a) bil-dik know-OP ‘acquaintance’(38b) oku-ma-dik read-neg-OP ‘unread’

(39) oku-ma-dik gazete kal-ma-diread-neg-OP newspaper stay-neg-di.past‘not-read-newspaper not stays’ = ‘No newspaper was left unread.’

Lewis (1967, 163) notices that the DIK-participles often show a passivemeaning, although they are constructed in the active form. This observation canbe stated more specifically as: The DIK-suffix changes verbal stems to passiveparticiples. In other words, it deletes another internal argument of the verbbeside the external argument. Lexicalized forms as in (38) and data from OldTurkish (750-1300 B.C.), where these participles are employed in passivefunction, confirm this point. Originally, all participles expressed an opendiathetic relation, i.e., they had either active or passive meaning (von Gabain1974, 76; 116). This can still be seen in certain uses of the future participle -EcEK: yiy-ecek bir hey al¦m ‘being-eaten something buy-I’, ‘I buy something toeat’. In the case of the DIK-participle, which primarily bears a perfectivemeaning, the passive use is lexicalized in Modern Turkish. In earlier times, theexternal argument (or subject) of the predicate used to be implicit to the

- 23 -

construction, while later it was expressed in the form of the possessive suffixcorresponding to the derivation of verbal nouns (cf. von Gabain 1974, 116).

The participle suffix -DIK can now be assigned a type that takes a transitiveVP and gets a participle of the type N/N: (NP\(NP\s))\(N/N) or TV\(N/x). Thoughin (35) the subject gazete bears no overt case ending, we assume an underlyingcase suffix for the analysis.

(39a) oku-ma-dik gazete -Ø kal-ma-diread-neg-OP newspaper-nom remain-neg-di.pastTV TV\(N/N)N N\NP NP\s

N/N N

NP s

‘No newspaper was left unread.’

This participle is rarely used in non-negated forms, since it contrasts with theactive participle -En formed from the corresponding passive verbal base:

(40) *kullan-d¦k vs. kullan-¦l-anuse -OP use -pas -SPTV TV\(N/x) TV TV\IV IV\(N/x)‘used’ (no record) ‘(being) used’

In the substantival use participles behave like verbal nouns. It is not clear fromthe literature why the attributive use of the participles has developed in adifferent direction than the nominal use of the infinitives (cf. Kissling 1960,179).

3.5 Verbal Nouns

Subordinate clauses are usually realized by nominalized sentences, which canfunction as subjects like in (44) or as objects with case endings in a matrixsentence like in (41-43). The internal arguments of the verb remain unchangedand stand in the canonical order. The subject of such embedded sentences caneither be identical with the subject of the matrix sentence as in (41), or it can bemarked with a possessive suffix on the nominalized verbal form. This isillustrated in (42) where the possessive suffix of the first person singular -(I)m

- 24 -

indicates the subject to the nominal okuma ‘reading’. This construction has beenextended to the izafet-construction, where the subject of the subordinate clauseis a genitive modifying the nominalized verb. Generally one can relate typeIV\(N) to the nominalization suffixes, i.e., they take an intransitive verb andchange it to a nominal, which then can be modified with the possessive, pluraland case suffixes. However, verbal nouns of this class can not be usedattributively.

(41) siz-i gör-me-ye gel-di-kyou.pl-acc see-NOM-dat come-di.past-1.pl‘We came to see you.’

(42) [bu yaz¦-y¦ oku-ma -m] -¦ söyle-dithis article-acc read-NOM -pos.1.sg -acc say-di.past‘he said the-my-reading-of-this-article’ =‘He told me to read this article.’

(43) [her yigid-in bir yogurt yiy-ih -i ]-Ø varevery young_man-gen a yoghurt-acc eat-NOM-pos.3.sg-nom there is

‘There is a manner to eat a yoghurt of every young man’‘Every young man has his own style to eat a yoghurt.’

The nominal use of the DIK participle is sometimes categorized as a verbalnoun (cf. Lewis 1967; Kissling 1960). However, it can only occur in thisfunction together with the possessive suffix: the form yaz-d¦g-¦m means either‘the fact that I wrote’ or ‘that which I wrote’. Similar to the case ofnominalization suffixes, the possessive suffix marks the subject, which receivesnominal subcategorization, i.e., the genitive. The nominalized sentence can takedifferent case suffixes according to its function in the matrix sentence:

(44) [Beatles’ler-in her yap-tigi]-Ø Inglitere’de moda olu-yorB.-pl-gen everything do-OP.3.sg-nomE.-loc fashion be-pres‘Everything the Beatles do is becoming the fashion in England.’

Here, the participial suffix -DIK is unified with the possessive suffix -I to buildup the unit -DIgI, which in general cannot be analyzed any further and is takenas a lexicalized nominalization suffix like -mE or -Ih. However, in the remainderof the discussion I show that this composed suffix can be given a reanalysis thatexplains the complex functional dependencies in relative clause constructions.

- 25 -

4. Turkish Relative Clauses

Turkish relative clauses are prenominal and marked by a relative suffix on thepredicate. There are two constructions for relative clauses, which are useddepending on whether the relativized element is the subject (or part of thesubject) or not (Lewis 1967; Underhill 1972; Hankamer & Knecht 1976).

If the head noun refers to the subject of the relative clause, the relativeconstruction is marked by a participial suffix of type ‘SP’: -En, -mih or -Er. Thesuffixes are assigned the type IV\(N/N): They take an intransitive verb and yieldan attributive adjective. The missing position in the sentence is then filled by thehead noun.

(45) ogul mekteb-e gid-erson school-dat go-aor.3.sg‘The son goes to school.’

(46) mekteb-e gid- en ogulschool-dat go- SP sonNP NP\IV IV\(N/N) N

IV N/N

N‘tgoing to school-son’ = ‘son that goes to school’

If the head noun refers to a position other than the subject position in therelative clause, the OP suffix -DIgI or -EcegI has to be chosen. The subject ofthe relative clause is expressed in the genitive which forms an izafet-construction with the participle. In (48), the participial suffix -DIgI is firstapplied to the transitive verb bekle yielding the nominalized form bekledigi.This form looks for an NP in the genitive, here kardeh-im-in, to form therelative clause kardeh-im-in bekle-dig-i, which modifies the head noun misafir.The OP suffix is assigned the complex functional type TV\(NP[gen]\(N/N)):

(47) kardeh-im misafir-i bekli-yorbrother-my guest-acc expect-pres‘My brother expects the guest.’

- 26 -

(48) kardeh-im-in bekle -digi misafirbrother-my-gen expect-OP.pos.3.sg guestNP[gen] TV TV\(NP[gen]\(N/N)) N NP[gen]\(N/N)

N/N N

‘guest that my brother expects’

This construction corresponds to the nominalized sentences discussed in section3.5. Although the suffix -DIgI seems to be an allomorph of the nominalizingsuffixes described there, it is a rather different morphological operator. It canform relative clauses, whereas the other nominalizing suffixes cannot. Thedifference between the -DIgI suffix and nominalizing suffixes becomes obviousunder a transparent reanalysis of the lexicalized complex consisting of the DIK-participle and the possessive suffix -I.

4.1 A Double Bracketing Paradox

Tracing back the functional structure of the OP-suffix -DIgI to the functionalstructure of its parts, a double bracketing paradox appears: The participial suffix-DIK was categorized above as TV\(N/N), which corresponds to itscharacterization as a morpheme forming a past perfect participle from atransitive verb. The possessive suffix -I obtained the category indexN\(NP[gen]\N) for marking the genitival modification on the head noun. Thedouble bracketing paradox arises through the application of the suffix -DIK tothe head noun crossing the possessive suffix -I, while the possessive suffixrefers back to the genitive: [1 kardeh-im-in bekle-[2 dig-i]1 misafir]2. Thus, thehead of this relative clause construction points in two directions. The functionalrelations in (49) are illustrated with arrows: The VP is changed to a nominal thatreceives its subject from the izafet-construction.

- 27 -

kardeh-im-in bekle -di g -i misafirNP[gen] TV TV\(N/N) N\(NP[gen]\N) N N/N

N

N[pos]

N

S

(49) N

NP[gen] VP OP pos N

This bracketing paradox can be solved with neither a harmonic nor adisharmonic functional composition. The two category indices in (50) can in noway be combined to yield the necessary category index TV\(NP\(N/N)) for thelexicalized suffix DIgI. The composition is only possible if we assume anunderspecified or polymorph index. As shown in section 3.1, all nominalsuffixes look for arguments of the polymorph type N/x, i.e., nouns or adjectives.Therefore, we suppose that the possessive suffix -I is of the underspecified type(N/x)\(NP[gen]\(N/x)). If it is applied to a nominal of type N and then applied toan NP in the genitive, it forms an expression of type N/x. In (51) the modifiedindices are combined with the help of the harmonic functional composition tothe index of the complex suffix -DIgI.

(50) DIg -I ⇒ DIgITV\(N/N)N\(NP[gen]\N) ⇒?? TV\(NP[gen]\(N/N))

(51) DIg- -I ⇒ DIgITV\(N/N) (N/x)\(NP[gen]\(N/x)) ⇒FCb TV\(NP[gen]\(N/x))

Using underspecified indices we can explain the relative clause constructions in(48) as well as the nominal sentences in (44). The difference between the simplenominalizing suffixes and the suffix -DIgI is expressed in the different typeassignments: nominalizing suffixes are of tye IV\N, whereas the suffix -DIgI

- 28 -

receives the polymorph type IV\(N/x). Only the latter index licenses themodifying use of a nominal as a relative clause.

4.2 The Bah¦bozuk or ‘Broken Head’ construction

The notion bah¦bozuk or ‘broken head’ construction was coined by Lewis (1967,259f.), who gave the following example for illustration: The genitive hu adam¦nof the subject bah¦ in (52) becomes the head noun of the relative clauseconstruction in (53). The possessive suffix -¦ on bah marks the reference of thehead noun to the genitive in the relative clause.

(52) [ [hu adam-¦n]NP[gen] bah-¦-Ø]NP[nom] bozukthis man-gen head-pos-nom confused‘The head of this man is confused.’

(53) bah-¦ bozuk adamhead-pos confused man ‘man, whose head is confused’

This construction can be transferred to verbal sentences having a finite verbform instead of the copula in (52). In this way the genitive adam¦n can be thehead of the relative clause k¦z¦ hukuk okuyan, which is formed with the SP-suffix -En. The possessive suffix on the subject of the relative clause marks themissing genitive, to which the head noun refers:

(54) adam-¦n k¦z-¦ hukuk oku-yorman-gen daughter-pos.3.sg law read-pres‘The daughter of the man studies law.’

(55) k¦z-¦ hukuk oku-yan adamdaughter-pos.3.sg law read-SPman‘man, whose daughter studies law’

This construction can be expanded to cases where the head noun refers to agenitive in a constituent, which is, contrary to (55), the subject of the relativeclause. In these cases the OP-suffix has to be used to form a relative clause:From sentence (56) the genitive adam¦n, which is located in the dative NPadam¦n mektebine, can be relativized. The genitive oglan¦n in (57) does notmodify the possessive NP mektebine, but the object gittigi:

- 29 -

(56) oglan adam-¦n mekteb-in-e gid-er son man-gen school-pos.3.sg-dat go-aor

‘The son goes to the school of the man.’

(57) oglan-¦n mekteb-in-e git- tigi adamson-gen school-pos.3.sg-dat go-OP man‘man, to whose school the son goes’

The general strategy as to what kind of relativation principle is used depends onthe constituent in which the genitive is located. For genitives that modifysubjects, the SP-participle is used, while for genitives that modify objects like in(57), the OP-suffix is chosen.

The analysis starts with two suggestions: (i) The choice of an appropriatesuffix depends on the number of unfilled argument slots of the predicate, asshown in section 4.1. (ii) The missing genitive, which is marked with thepossessive suffix, can be ‘passed on’ using functional composition, until, onthe level of the argument places, it influences the selection of the suffix. In(55a) only one argument is missing and accordingly the SP-suffix isselected, while in (57a) two arguments are missing, which governs theselection of the OP-suffix. The subject oglan¦n of the relative clause (57a)modifies the participial expression mekteb-in-e git-tigi.

(55a) k¦z- ¦- Ø hukuk oku-yan adamdaughter-pos.3.sg study law-SP manN N\(NP[gen]\N) N\NP NP\s (NP\s)/N/N N

NP[gen]\N FCb

NP[gen]\NP FCb

NP[gen]\s FCb

N/N N

‘man, whose daughter studies law’

- 30 -

(57a) oglan-¦n mekteb-in -e git -tigi adamson-gen school-pos.3.sg- dat go -OP manNP[gen] N N\(NP[gen]\N) N\NP FCb TV TV\(NP[gen]\(N/N)) N

NP[gen]\NP NP\(NP\s) FCb

(NP[gen]\NP)\s NP[gen]\(N/N)

N/N N

‘man, to whose school the son goes’

4.3 Flat Reconstruction of Embedded Sentences

The examples of the last section showed that the choice of a relative clauseconstruction does not depend on the function of the expression to which thehead noun refers, but is indeed dependent on the function of the co-constituentof the predicate to which the head noun refers. Turkish seems to be ‘blind’ to adeeper hierarchical structure; rather it operates on the highest syntactic level.Whether this is the case for other grammatical phenomena as well will needfurther investigation.

Finally, the embedding of relativized elements in subject sentences is brieflydiscussed (cf. Hankamer & Knecht 1976, 127ff.). If an element of a subjectsentence is relativized, the corresponding strategy is oriented towards thefunction of the sentence regarding the matrix predicate: The nominal sentencey¦lan¦n kabag¦ yedigi in (58) is the subject of the predicate hüpheli(-dir). If thegenitive y¦lan¦n, which is the subject of the nominal sentence, is relativized as in(59), the SP-participle olan is selected as copula.

(58) [y¦lan-¦n [kabag-¦]NP[acc]ye-digi-Ø]NP[nom] hüpheli(-dir)snake-gen squash-acc eat-OP-nom doubtful(copula)‘That the snake ate the squash is doubtful.’

(59) [ [kabag-¦]NP[acc] ye-digi-Ø]NP[nom] hüpheli ol-an y¦lansquash-acc eat-OP-nom doubtful be-SP snake‘snake of which it is doubtful whether it ate the squash’

However, if a genitive that is subject of an object sentence is relativized, theOP-strategy has to be employed. In (60) the verb san¦yor governs the objectsentence y¦lan¦n kabag¦ yedigini ‘the snake’s eating of the squash’, where thegenitive y¦lan¦n is the subject. If this genitive is relativized like in (61), the

- 31 -

whole sentence is nominalized to a participial form, which is formed by the OP-suffix DIgI. In (61), the genitive Hasan¦n is the subject of sand¦g¦, while therelativized genitive is the subject of yedigini.

(60) Hasan [y¦lan-¦n [kabag-¦]NP[acc] ye-digi-ni]NP[acc] san¦-yorHasan-nom snake-gen squash-acc eat-OP-acc believe-pres‘Hasan believes that the snake ate the squash.’

(61) [Hasan-¦n [ [kabag-¦]NP[acc] ye-digi-ni]NP[acc] san-d¦g¦ y¦lan[Hasan-gen [squash-acc eat-OP]-acc believe-OP] snake‘snake, of which Hasan believes that it ate the squash’

To summarize, we can state that the mechanism of generalized categorialgrammars is able to analyze the complex functional relations of Turkish relativeclause constructions. The function of the complex suffix -DIgI has beencomposed from the functions of the participial suffix -DIK and the possessivesuffix -I. With this the difference to other nominalizational suffixes has beenreconstructed. It has been shown on embedded genitives that Turkish doesn’temploy a hierarchical structure in the selection of the relative clauseconstructions, but uses a ‘flat’ analysis, a fact that has been reconstructed verywell by using the categorial grammatical mechanisms.

Bibliography

Ajdukiewicz, K. 1935. Die syntaktische Konnexität, Studia Philosphica 1, 1-27.Bar-Hillel, Y. 1953. A quasi-arithmetical notation for syntactic description.

Language, vol 29, 47-58.Buszkowski, W. & Witold, M. & van Benthem, J. (eds.) 1988. Categorial

Grammar. Amsterdam; Philadelphia: Benjamins.Cresswell, M. 1973. Logics and Languages. London: Methuen.Di Sciullo, A. & Williams, E. 1987. On the Definition of Words.

Cambridge/Mass.: MIT Press.Erguvanl¦, E. 1984. The Function of Word Order in Turkish Grammar.

Berkerley; Los Angeles; London: University of California Press.von Gabain, A. 1974. Alttürkische Grammatik. 3. Aufl. Wiesbaden:

Harrassowitz.

- 32 -

Geach, P. 1972. A Program for Syntax. In: D. Davidson & G. Harman (eds.).Semantics of Natural Language. Dordrecht: Reidel, 483-497.

Haddock, N. & Klein, E. & Morrill, G. (eds.) 1987. Working Papers inCognitive Science, Vol. 1: Categorial Grammar, Unification Grammar andParsing. University of Edinburgh: Center for Cognitive Science.

Hankamer, J. & Knecht, L. 1976. The Role of the Subject/Non-SubjectDistinction in Determining the Choice of Relatice Clause Participle inTurkish. Montreal Working Papers in Linguistics 6, 123-135.

Kissling, H. 1960. Osmanisch-Türkische Grammatik. Wiesbaden: Harrassowitz.Klein, E. & van Benthem, J. 1988. Categories, Polymorphism and Unification.

Institute for Cognitive Science (University of Edinburgh)/ Institute forLanguage, Logic and Information (University of Amsterdam).

Lambek, J. [1958] 1988. Categorial and Categorical Grammars. In: R. Oehrle, E.Bach & D. Wheeler (eds.). Categorial Grammars and Natural LanguageStructures. Dordrecht: Reidel, 297-318.

Lesniewski, S. 1929. Grundzüge eines neuen Systems der Grundlagen derMathematik. In Fundamenta Mathematicae 14, 1-81.

Lewis, G. L. 1967. Turkish Grammar. Oxford: Clarendon.Montague, R. 1974. Formal Philosophy. Selected Papers of Richard Montague.

Ed. R. H. Thomason. New Haven: Yale University Press.Moortgat, M. 1988. Mixed composition and discontinuous dependencies. In: R.

Oehrle, E. Bach & D. Wheeler (eds.). Categorial Grammars and NaturalLanguage Structures. Dordrecht: Reidel, 319-348.

Oehrle, R., Bach E., & Wheeler, D. (eds.) 1988. Categorial Grammars andNatural Language Structures. Dordrecht: Reidel.

Selkirk, E. 1982. The Syntax of Words. Cambridge/Mass.: MIT Press.Underhill, R. 1972. Turkish Participles. Linguistic Inquiry 3, 87-99.Underhill, R. 1986. Turkish. In: D. Slobin & K. Zimmer (eds.). Studies in

Turkish Linguistics. Amsterdam: Benjamins, 7-21.Uszkoreit, H. 1986. Categorial Unification Grammars. Artificial Intelligence

Center at SRI and CSLIWilliams, E. 1981. On the Notions ‘Lexically Related’ and ‘Head of Word’.

Linguistic Inquiry 15, 639-673.